The distinct generative models
Let's think about the distinct methods used for generative modeling.
First, let's list some works that relates to drifting models, the novel method for generative modeling.
Diffusion models, generative adversarial networks, moment matching, and contrastive learning.
Diffusion models are neural networks that, at inference time, take a noise sample and transform it into an image.
Generative adversarial networks use stwo neural networks with adversarial optimization. There is not multi-steps; it is single-step generation.
Moment matching aims to minimize the maximum mean discrepancy between the generated and data distributions.